Canva 发布首款自研图像模型 Canva Craft，生成前先规划设计
Canva’s first AI image model plans designs before generating them
- 1Canva 发布首款自研图像生成模型 Canva Craft，称直接对标 OpenAI、Microsoft 和 Google 的同类模型。
- 2官方称两阶段架构先规划设计再生成图像，更擅长清晰文字和精确物体位置，精度、一致性与可控性更好。
- 3模型最高支持 2K 分辨率和 30 多种宽高比，已面向全球所有用户开放，入口在 Dream Lab。
- 4在众包基准 Design Arena 排行榜上排名第四，落后于 OpenAI 三款 GPT-Image 系列模型。
Canva 自研图像模型主打设计场景的精度与可控性，先规划再生成是差异化路线，榜单排名仍落后 OpenAI 产品。
- 影响：营销与设计团队可在常用设计工具内直接生成素材，或减少单独图像工具的开支；该模型在独立基准上尚落后 OpenAI 产品，替换现有工具前需按自身场景评测；后续 MCP 接入为智能体参与设计流程留出空间。
- 参考：可考虑让设计团队用同一批含清晰文字和固定排版的素材任务，对比 Canva Craft 与现有图像工具的产出；已在用 Canva 的团队可在 Dream Lab 小范围试用，再决定是否调整图像工具预算。
- 依据：两阶段架构、训练数据与安全合规措施为 Canva 官方口径，暂无独立验证；排名第四来自众包基准 Design Arena。
Canva is diving even deeper into AI with the launch of an image generation model that competes directly with those from OpenAI, Microsoft, and Google. The model, called Canva Craft, uses a two-stage architecture that allows it to plan a design before generating an image, allowing for “greater precision, consistency and control,” according to an announcement on Thursday.
Canva Craft can generate images at up to 2K resolution and in more than 30 aspect ratios. In addition to generating social media posts and marketing content, it’s also supposed to create realistic scenes, digital art, environmental imagery, and more. The model’s two-stage system allows it to more accurately generate complex images that require legible text and precise object placement, according to Canva.
1/5
The new AI model builds on Canva’s big push into AI. In April, Canva launched new prompt-based editing capabilities that allow you to create or make changes to a design by describing it. The design tool also brought layers to AI-generated images, which let you make changes to part of an image while leaving the rest untouched.
Canva says it trained Craft on “millions of human-made templates” from the platform’s content library, which includes things like social media posts, presentations, infographics, and menus. The model is currently ranked fourth on Design Arena’s leaderboard for the best AI image generation models, a crowdsourced benchmark for AI designs. It sits behind OpenAI’s GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare, and GPT-Image-2.
Canva notes that its AI model comes with “robust input and output moderation, privacy and safety controls, and reporting tools.” The company plans to allow AI agents to plug into Canva Craft through the open Model Context Protocol, and will also roll out more efficient Craft-fast and Craft-ultrafast models in the future.
Canva Craft is available now to all users globally. You can try it by navigating to the three-dot option on Canva’s left-hand menu and then selecting Dream Lab.
Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.
- Emma Roth
来源：The Verge · AI · theverge.com